Strange Node Assert Failure In Github Actions

[drew-y]

There is a single very simple test failing in my Github Actions PR Check. The test performs an assert.deepStrictEqual on two straight forward object data structures.

Here's the thing. I cannot for the life of me reproduce this failure outside of github actions.

This whole suite passes on my local machine. It even passes when I run the github workflow locally in docker using act and the command act pull_request --container-architecture linux/amd64.

Even stranger, the reported diff of the failing test (expected, actual) is identical.

The runner and my machine are using the exact same node and npm versions

node: v22.7.0
npm: 10.8.2

I've also tried different NODE_ENV flags to no avail.

Has anyone else run into this? Am I missing something obvious?

[drew-y]

I've also double checked that there are no caches causing strange behavior.

[drew-y]

I understand that a diff can sometimes not account for object prototypes. I don't think that applies here, but I'll do a fresh JSON parse on BOTH objects to be sure.

Still if thats the case, why does it specifically call out a subsection of the data structure that is identical?

Update: A fresh stringify then parse made no difference.

[cjihrig]

I don't think this is related to the test runner. This appears to be related to the node:assert module.

Also, the results don't appear to be identical. The failed assertion prints out the actual and expected values. You can copy them locally and diff them to see that they are different.

As to why it happens in GitHub Actions - I have no idea.

[drew-y]

Interesting about the diff command. Can confirm it shows a diff for me there too.

[drew-y]

As far as I can tell @cjihrig those are just whitespace differences. Ignoring whitespace, I get:

❯ diff -b a.txt b.txt                                                                                                  (base) 
1c1
<       expected:
---
>       actual:

I'm not sure how white would matter here since the comparison is between objects, and not the stringified versions.

[drew-y]

I printed the JSON of each object for comparison. There is indeed a difference in the count of arrays for some reason.

[cjihrig]

I think the difference in whitespace probably indicates a difference in nesting. In other words, maybe the objects are being inserted into the wrong array. I can't say for sure because I'm unfamiliar with your code.

[drew-y]

Thats the issue. I've narrowed the problem down to something on my end (of course 🤦🏻 ). I accidentally snuck in a change that ties nesting structure to the folder structure of the repo. Which is probably different in the Github runner.

I'm a bit surprised that act didn't catch the issue. I guess its an really imperfect representation (at least in how I'm using it).

I really appreciate the help @cjihrig.

[drew-y]

TLDR: Test really was accurately failing. Test was tied to folder structure that was different in the Github action runner. The act command did not catch this, adding to the confusion.. Solution was to fix the folder structure pointer in the test.